diff options
| author | Michael Butler <butlermichael@google.com> | 2018-02-13 17:25:57 -0800 |
|---|---|---|
| committer | Michael Butler <butlermichael@google.com> | 2018-02-20 19:13:06 -0800 |
| commit | 084401d6215dca122999261c5ac3718ebf61b29e (patch) | |
| tree | fd1afdd052d4dcecf5103bd7e90333981bad2502 /nn/runtime/include/NeuralNetworks.h | |
| parent | e78b89d46b7e534b749ace863886571a840e83e2 (diff) | |
| download | ml-084401d6215dca122999261c5ac3718ebf61b29e.tar.gz | |
NNAPI narrow evaluation for P -- runtime
We have determined that for Android P it is sufficient to have a
mechanism for a developer to specify on a per-model basis that it is
acceptable for FLOAT32 operations to be carried out as if they were
FLOAT16 operations. This CL manages the versioning differences between
1.0 and 1.1.
Bug: 63911257
Test: mm
Test: NeuralNetworksTest
Test: VtsHalNeuralnetworksV1_0TargetTest
Change-Id: If6f31536eedc92c4795056fdf3ff8818db1bc988
Diffstat (limited to 'nn/runtime/include/NeuralNetworks.h')
| -rw-r--r-- | nn/runtime/include/NeuralNetworks.h | 22 |
1 files changed, 22 insertions, 0 deletions
diff --git a/nn/runtime/include/NeuralNetworks.h b/nn/runtime/include/NeuralNetworks.h index 00a60d309..008c1ab50 100644 --- a/nn/runtime/include/NeuralNetworks.h +++ b/nn/runtime/include/NeuralNetworks.h @@ -1953,6 +1953,28 @@ int ANeuralNetworksModel_identifyInputsAndOutputs(ANeuralNetworksModel* model, u const uint32_t* outputs); /** + * Specifies whether {@link ANEURALNETWORKS_TENSOR_FLOAT32} is allowed to be + * calculated with range and/or precision as low as that of the IEEE 754 16-bit + * floating-point format. By default, {@link ANEURALNETWORKS_TENSOR_FLOAT32} + * must be calculated using at least the range and precision of the IEEE 754 + * 32-bit floating-point format. + * + * @param model The model to be modified. + * @param allow 'true' indicates {@link ANEURALNETWORKS_TENSOR_FLOAT32} may be + * calculated with range and/or precision as low as that of the + * IEEE 754 16-bit floating point format. 'false' indicates + * {@link ANEURALNETWORKS_TENSOR_FLOAT32} must be calculated using + * at least the range and precision of the IEEE 754 32-bit floating + * point format. + * + * Attempting to modify a model once {@link ANeuralNetworksModel_finish} has been + * called will return an error. + * + * See {@link ANeuralNetworksModel} for information on multithreaded usage. + */ +int ANeuralNetworksModel_relaxComputationFloat32toFloat16(ANeuralNetworksModel* model, bool allow); + +/** * Create a {@link ANeuralNetworksCompilation} to compile the given model. * * <p>This only creates the object. Compilation is only performed once |